Urdu Katib Handwritten Dataset: A Historical Document Dataset for Offline Urdu Handwritten Text Recognition with CRNN-Based Baseline Evaluation

2026-06-17 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

The Urdu Katib Handwritten Dataset (UKHD) is a specialized real dataset designed to advance research in offline Urdu Handwritten Text Recognition (UHTR), a field with limited prior research due to unique script challenges and data scarcity. This dataset is the first offline Urdu handwritten text lines collection specifically curated from historical materials written by Katibs, encompassing diverse flat nib writing variations in the Nastalique calligraphic style. The study also evaluated various CRNN-based hybrid models to identify optimal architectures for Urdu Katib Handwriting Recognition (UKHR). The CNN-BGRU-CTC model demonstrated robust performance, achieving low Character Error Rate (CER) and Word Error Rate (WER), aiming to support the development of robust recognition systems for preserving Urdu handwritten literature.

Key takeaway

For Machine Learning Engineers developing Urdu Handwritten Text Recognition systems, the new Urdu Katib Handwritten Dataset (UKHD) provides a crucial resource. You should leverage this dataset to train and benchmark models, particularly considering the CNN-BGRU-CTC architecture which demonstrated robust performance with low Character Error Rate and Word Error Rate, to advance the preservation of Urdu handwritten literature.

Key insights

The Urdu Katib Handwritten Dataset (UKHD) addresses a critical gap in resources for offline Urdu Handwritten Text Recognition.

Principles

• Cursive script HTR faces unique challenges due to script complexity and data scarcity.
• Benchmark datasets are crucial for advancing research in under-resourced languages.

Method

Evaluate CRNN-based hybrid models to identify optimal architectures for specific cursive scripts like Urdu Nastalique, focusing on metrics like Character Error Rate and Word Error Rate.

In practice

• Utilize the UKHD for training and benchmarking Urdu HTR models.
• Consider the CNN-BGRU-CTC model as a strong baseline for Urdu Katib Handwriting Recognition.

Topics

• Urdu Handwritten Text Recognition
• Historical Documents
• Nastalique Calligraphy
• CRNN Models
• Dataset Development
• Character Error Rate

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• A Text Recognition Dataset from Sahidic Coptic Ancient Manuscripts — A new dataset, SCAM, highlights significant challenges for Handwritten Text Recognition in low-resource historical scripts like Sahidic Coptic.
• BullingerDB: A Dataset for Handwritten Text Recognition and Writer Retrieval — BullingerDB is a large, multilingual dataset for historical HTR and writer retrieval, introducing temporal metrics and highlighting challenges from long-term stylistic variation.
• UrduMMLU: A Massive Multitask Benchmark for Urdu Language Understanding — Native, culturally-grounded benchmarks reveal significant LLM knowledge gaps in non-English, humanities-focused content.
• Koshur Diacritizer: A Byte-Level Sequence-to-Sequence Model for Kashmiri Diacritic Restoration — Koshur Diacritizer is a ByT5-small model that restores Kashmiri diacritics using a new 23.7k sentence pair dataset.
• Curation of a Palaeohispanic Dataset for Machine Learning — A new structured dataset enables machine learning analysis of under-resourced Palaeohispanic languages.
• Released a free 9.8M doc Indic multilingual corpus — Hindi, Bengali, Tamil, Telugu + 7 more (CC0, HuggingFace) [P] — A new 9.8M document, 11-language Indic multilingual corpus is now freely available under CC0.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.